Monitoring with uncertainty

TL;DR

This paper addresses runtime verification challenges when monitoring gaps occur due to overhead control, proposing statistical models to learn application behavior and estimate property violations despite incomplete traces.

Contribution

It introduces methods to use statistical models for filling gaps and estimating violation probabilities in incomplete runtime traces.

Findings

Statistical models can effectively learn application behavior.

Techniques can estimate violation probabilities with incomplete data.

Methods improve reliability of runtime verification under monitoring constraints.

Abstract

We discuss the problem of runtime verification of an instrumented program that misses to emit and to monitor some events. These gaps can occur when a monitoring overhead control mechanism is introduced to disable the monitor of an application with real-time constraints. We show how to use statistical models to learn the application behavior and to "fill in" the introduced gaps. Finally, we present and discuss some techniques developed in the last three years to estimate the probability that a property of interest is violated in the presence of an incomplete trace.